With 10 hours of conversation data and one hour of recording, Saltlux will recreate an MC and AI assistant who can understand and communicate fluently in Vietnamese.
Saltlux's AI Assistant and Metahuman (recreating real people in virtual space) are among the models that have attracted great attention from visitors to the AI Expo exhibition within the framework of the Vietnam Artificial Intelligence Festival ( AI4VN 2023 ), taking place in Ho Chi Minh City.
AI assistants can work similarly to popular chatbots like ChatGPT but can communicate in Vietnamese flexibly and accurately in many contexts. Meanwhile, Metahuman makes many people excited when creating a copy of any person in the digital environment.
These two models can be combined together to create an interesting MC AI version that can communicate in Vietnamese with information updated in real time.
Visitors experience the AI Assistant and Metahuman models at the AI4VN Exhibition held on September 21 in Ho Chi Minh City. Photo: Khuong Nha
Saltlux's technology allows a person to create a copy of themselves after an hour of recording and about 10 hours of conversation. After collecting enough image data, the AI will automatically copy the images, expressions, movements and intonation of a real person, simulate the form of a virtual MC and put it into use in many contexts.
To accurately describe each person's facial features and expressions, the developer has integrated many knowledge graph recognition technologies, allowing semantic segmentation, pose estimation, facial recognition, hand gestures, age group classification, gender... so that the virtual MC version looks almost like the real one, and ordinary users can hardly recognize the difference.
Next, the voice analysis technology will process the natural language to accurately reproduce the real person's voice and intonation before being inserted into the virtual MC model. Saltlux uses the Hybrid-Tacotron deep learning model, allowing the AI to reproduce both the mid-range and high-pitched voices of real people, ensuring maximum stability and personalization.
Then, the speech synthesis engine that allows text to speech and vice versa - speech to text - will allow the AI MC and the person to interact directly in Vietnamese smoothly.
In real-life experiences in noisy conditions, Saltlux's MC AI still fully hears the lines spoken directly at the scene and converts them into text, before executing the given commands. AI recognizes accents from different regions well and responds quickly to content when users suddenly change the topic. Users can interact with AI by voice, drag and drop commands or type content into the chat box.
One of the advantages of this Vietnamese-speaking AI MC is the accuracy of the answers. Instead of thinking of answers like ChatGPT, Saltlux maximizes the control of the answers by dividing each field. For example, with current news, AI will only get news from reputable newspapers. After synthesizing and giving results, AI will cite the articles so that users can directly access the link if they want to read in depth.
When asking questions related to a tourist destination, the developer will prepare content in advance, put it into many different contexts to predict the user's question, thereby providing appropriate answers with high accuracy. Dividing each content area and being flexible in creating input and output filters helps significantly reduce the "illusion" of AI chatbots' self-created answers.
In Korea, this AI model has created the "Goodpy National Assistant", which is used by the government to provide public administrative services to 20 million people. In Vietnam, the AI model and virtual MC were researched and operated by Saltlux two years ago. One of the outstanding applications is the virtual tour guide at the Tam Chuc Pagoda relic complex. The system not only helps visitors answer questions, guide, give directions, and explain about each location, but can also accurately determine each person's location in real time, thereby providing appropriate instructions.
The Ploonet Studio platform interface allows users to create content with AI without the need for powerful devices. Photo: Khuong Nha
Once you have a virtual MC and voice, users can customize the content and hosting style on the Ploonet Studio platform to mass produce videos.
The Ploonet Studio interface is also simple, users just need to click and wait about two minutes, a new video will be published by the system. Everything is done on the cloud server so there is no requirement for terminal configuration. At CES 2023 in Las Vegas, the Ploonet Studio platform was honored with the Innovation Award.
Saltlux representative said that their AI model helps improve work efficiency by 20% and cut operating costs by more than 30%, ensuring high availability 24/7.
To operate this MC AI model, users need to pay two fees. One is a fixed fee for recreating real people in virtual space. The price of the package will vary depending on the difficulty requested by the customer. The second fee is a fee for creating videos on Ploonet Studio, which is charged by the hour. Saltlux has not disclosed specific prices for individual customers. The company is targeting the group of business customers and organizations, governments.
Vnexpress.net
![[Photo] Prime Minister Pham Minh Chinh chairs Government Conference with localities on economic growth](https://vstatic.vietnam.vn/vietnam/resource/IMAGE/2025/2/21/f34583484f2643a2a2b72168a0d64baa)
Comment (0)